Qwen3.8-27B EXL3 4bpw on Arc Pro B70 (exl3xpu) — 2 results + catalog entries - #49
Merged
Merged
Conversation
… Pro B70 Catalog: new runtime exl3xpu (0xSero/exl3xpu, the ExLlamaV3 XPU port) and quant exl3-4bpw, linked to the qwen3-8-27b board; seed migration included. Results: two measured runs from campaign CAM-2026-AUTOROUND-VS-EXL3 — 150 W (median decode 17.1 tok/s post-first stream rate, n=9-10) and 230 W stock cap (24.6). fp8 KV + native MTP3; post-first client rate is a conservative lower bound under speculative decode. 262,144-token native context proven on one card; details and caveats in the notes fields.
|
Someone is attempting to deploy a commit to the Community Labs Team on Vercel. A member of the Team first needs to authorize it. |
- results/SergiioB/2026-09-27-qwen3-8-27b-exl3-4bpw-exl3xpu-230w.json: imported (result 93) - results/SergiioB/2026-09-27-qwen3-8-27b-exl3-4bpw-exl3xpu.json: imported (result 94) Merge ingestion completed. Rerunning this workflow will not duplicate these results. |
Contributor
Author
|
Trimmed to result files only. The catalog entries (runtime |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds the first EXL3 results for Qwen3.8-27B on Intel Arc Pro B70:
results/SergiioB/2026-09-27-…-exl3xpu.json— 150 W cap: decode 17.1 tok/s (client post-first stream rate, median of 9–10 samples, p512/g128), cold 8K prefill ~1,680 tok/s, TTFT ~440 ms, ctx 65,536…-230w.json— stock 230 W cap: sustained 24.6 tok/s (+33–42%), ctx 65,536Because
exl3xpuandexl3-4bpwwere not in the catalog, this PR also adds:RUNTIMES:exl3xpu→ https://github.com/0xSero/exl3xpu (ExLlamaV3 XPU port; serves via its own vLLM-XPU-derived API surface)QUANTS:exl3-4bpw(EXL3 trellis, ~4 bpw) + theqwen3-8-27bboard linksupabase/migrations/20260927120000_catalog_exl3_4bpw_exl3xpu.sqlmirroring the pattern of earlier catalog migrationsMeasurement honesty notes (in the files too)
Validation
npm run results:validate— both files pass against the updated catalognpm run catalog:validate— 46 hw / 32 models / 18 quants / 8 runtimes, all resolveIngestion may need the catalog migration deployed before the result check passes — happy to split into two PRs if preferred.